V-Ray 2.0: VRayBPTracer

[vlado]

Yes, I have read it, I did experiments with it and I came to the conclusion that MLT is way overrated. It can be very useful in some special situations, but for most everyday scenarios, it performs (much) worse than a well-implemented path tracer. This is because MLT cannot take advantage of any sort of sample ordering (e.g. quasi-Monte Carlo sampling, or the Schlick sequence that we use, or N-rooks sampling etc). A MLT renderer must fall back to pure random numbers which greatly increases the noise for many simple scenes (like an open skylight scene).

Best regards,
Vlado

[stevesideas]

Can someone explain to me in Layman's terms (as Im not that much of a techy guy) how the BPTracer differs from normal GI ? (If there is any correlation at all - maybe ive misunderstood the whole thing and its just to do with caustics)

[anchovy]

In short the BPT, PPT MLT, lightcache, photon mapping and other techniques are all trying to solve the radiance equation, what we sometimes refer to as GI.

Path tracing, Bidirectional path tracing and Metropolis light transport if set up correctly try to do this through brute force calculation of how light propagates through the scene. They each, given enough time, can theoretically arrive at a "correct"solution. The trade off is, they take time, sometimes a lot! and can also end up with noise, a product of the variance between adjacent samples.

Techniques like Cornell's radiosity, Photon mapping, various caching schemes are generally faster but are prone to errors due to the fact that they are taking sparse samples of what the solution looks like and try to fill in the blanks.

As to how specifically BPT is different to "normal" GI it really depends what solver you are comparing it to.

http://en.wikipedia.org/wiki/Path_tracing

Hope that helps!

-Michael

[anchovy]

Yes, I have read it, I did experiments with it and I came to the conclusion that MLT is way overrated. It can be very useful in some special situations, but for most everyday scenarios, it performs (much) worse than a well-implemented path tracer. This is because MLT cannot take advantage of any sort of sample ordering (e.g. quasi-Monte Carlo sampling, or the Schlick sequence that we use, or N-rooks sampling etc). A MLT renderer must fall back to pure random numbers which greatly increases the noise for many simple scenes (like an open skylight scene).

Best regards,
Vlado

Vlado-

Thanks for the info there. So glad youve looked at it. I certainly understand there are some downsides, time/offering a clean solution being the big challenges. Seems like before too mayy years we may see solutions solely based on brute force. I like the elagance in theory of capturing some of the more subtle fx without explicitly using various techniques to arrive at them. And this coming form a film guy vs visualization. I know the value of cheating. We certainly live in interesting times.

http://raytracey.blogspot.com/2010/1...nsport-on.html
http://www.youtube.com/watch?v=3Xo0qVT3nxg
http://www.youtube.com/watch?v=70uNj...eature=related

-Michael

[stevesideas]

Thanks. I'll do some reading on this too.

[anchovy]

http://morphographic.com/PathTrace/buddha_Test_02.rar

Vlado-

I took a stab at a tough scene for the PPT to resolve. You might recognize the layout from The Mlt paper. You can just begin to see the caustics forming at the base if the curved/faceted mirror. Id love to see what the BPT can do with it.

Best of luck!

-Michael

[jacksc02]

Im doing a 1000 pass test with BPT at the moment on your scene
Stay tuned....

[jacksc02]

Here is a BPT render which took 58mins and did 1000 passes

[anchovy]

Hey Chris that looks great, noisy but great to see the caustics resolving after only 1000subdivs. Any chance yo can run one with more like 20k?

Thanks!

-Michael

[vlado]

Directly visible caustics are easy; it's caustics in mirrors or through glass that are a problem for unbiased engines

Best regards,
Vlado

[anchovy]

Vlado-

Any thoughts as to what it might take to get this scene clean? It sounds like you are working on reducing the variance in general.

-Michael

[jacksc02]

Anchovy-
Here is a 20k subdivs render.
Took 19 hours and is still noisy in the mirror.

[vlado]

Vlado-

Any thoughts as to what it might take to get this scene clean? It sounds like you are working on reducing the variance in general.

-Michael

Well, with this scene, most of the commercial "unbiased" renders will simply give you a filtered denoised image.

Other than that, there are several things that we are working on, but I don't know yet which will work best.

Best regards,
Vlado

[dlparisi]

OT: What scene is that in Vlado's third render (the first post)? I used to have it but can't locate it now.

[vlado]

It was from some render challenge a few years back...

Best regards,
Vlado